CSE Seminar
Machine Learning and Causality: Building Efficient, Reliable Models for Decision-Making
This event is free and open to the publicAdd to Google Calendar
Passcode: 445475
President’s Postdoctoral Fellowship Program Candidate
Abstract:
Increasingly, practitioners are turning to ML methods to provide guidance in key decision making processes. However, making decisions based on incidental correlations rather than causal effects can lead to adverse, or unfair outcomes. In this context, it is hard to overestimate the importance of training models that learn causal relationships that can be used to guide personalized interventions.
In this talk, I will present my work that addresses inefficiencies in causal learning for decision making. The majority of causal literature focuses on learning accurate individual treatment effects, which may be complex and hard to estimate from small samples. However, it is often sufficient for the decision maker to have estimates of upper and lower bounds on the potential outcomes of decision alternatives to assess risks and benefits. In my work, I show that in such cases we can improve sample efficiency by estimating simple functions that bound these outcomes instead of estimating their conditional expectations. My analysis highlights a trade-off between the complexity of the learning task and the confidence with which the learned bounds hold. I will present a novel algorithm that leverages these theoretical insights, learning reliable upper and lower bounds on potential outcomes. Using a clinical dataset and a well-known causality benchmark, I will demonstrate that my proposed algorithm outperforms baselines, providing tighter, more reliable bounds.
Bio: Maggie Makar is a PhD student advised by John Guttag at CSAIL, MIT. Maggie studies machine learning, causal inference, and the intersection between the two. Her work focuses on building data-efficient causal inference methods in resource-constrained settings, and building robust predictive ML models using ideas from causality. Her research is characterized by a dual focus on advancing both the fundamental science of statistical learning methods, and making progress on real world problems, especially in healthcare. Prior to MIT, Maggie received a B.Sc. in Math and Economics from the University of Massachusetts, Amherst.